Peptides play a decisive role in many physiological processes and as a result are playing an increasing role in the development of vaccines and peptide, peptidomimetic, and small-molecule drugs. Because of an explosion of functional and structural-genomic data, there is an urgent need for new methods to analyze and predict peptide-protein interactions, to allow this data to be effectively distilled into drugs and vaccines. In this proposal, we describe a new solution to this problem, through development of a new approach to describe and predict peptide-protein interactions for structurally solved proteins using Markov Random Fields (MRF). Free energy minimization of the MRF yields a probability distribution called a 3D probabilistic peptide profile or 3D profile. The 3D profile probabilistically specifies types, locations, orientations, and conformations of amino acids within active sites that can be connected to form energetically favorable, preferably long, polypeptide chains. 3D profiles can then be used to (a) recognize peptides that will bind, or to (b) generate optimized combinatorial libraries of peptides for testing. MRF models incorporate detailed energetic information and can incorporate prior knowledge on the target system including (i) sequences of peptides known to bind; (ii) structurally determined peptide/protein complexes; (iii) protein active site mutagenic information; and (iv) NMR-derived distance constraints. Multiple MRF models can be combined to account for protein flexibility. MRF models are created by initially positioning amino-acid probes into a fine grid in the active site. Fast Belief Propagation methods then minimize the internal MRF free energy, by optimizing beliefs for specific amino acids at specific active site positions while adjusting their positions and orientations. Final peptide conformations and libraries are obtained by marginalizing the profile. The MRF approach is novel and has significant principled advantages over existing methods that docking individual peptides to a target. A robust software prototype has been implemented; initial results are given for a PDZ domain. In Phase I, we will complete the prototype and apply it to SH2/SH3 domains, PDZ domains, and MHC l/ll domains. In Phase II, we will optimize and utilize the methods to tackle problems of pharmaceutical and biodefense interest that may include development of substrate-competitive inhibitors to kinases or inhibitors of YopH, a Yersinia Pestis protein tyrosine phosphatase. [unreadable] [unreadable]